Scientific methods
Data provenance, numerical assumptions and evidence are tied to model and task versions.
Data and model
The registered graph aggregates source weights into directed neuron pairs. Computation uses a versioned integer point LIF model with explicit leak, threshold, reset, refractory and synaptic propagation rules.
Each task records its initial state, background input and neurotransmitter-to-sign mapping. These declared numerical assumptions make paired control and stimulus responses reproducible.
Task readouts define the target population, the sampled off-target population and observation windows. Pass rates and model stimulus costs characterize that task; physiological prediction is evaluated separately against licensed external observations.
Evidence levels
Evidence is recorded by method: complete-model computation, protocol replay, reproduction across implementations and evaluation against external observations. Each result identifies the method, version and scientific scope it supports.
External model evaluation
An executable import and evaluation pipeline records source, license, checksums, calibration and held-out splits, baselines and sensitivity runs. External observation data are required before accuracy or generalization can be reported.
Evaluation import format ↗Evaluation tool ↗